Modeling Peripheral Olfactory Coding in Drosophila Larvae Derek J. Hoare ¤ , James Humble, Ding Jin, Niall Gilding, Rasmus Petersen, Matthew Cobb, Catherine McCrohan* Faculty of Life Sciences, The University of Manchester, Manchester, United Kingdom Abstract The Drosophila larva possesses just 21 unique and identifiable pairs of olfactory sensory neurons (OSNs), enabling investigation of the contribution of individual OSN classes to the peripheral olfactory code. We combined electrophysiological and computational modeling to explore the nature of the peripheral olfactory code in situ. We recorded firing responses of 19/21 OSNs to a panel of 19 odors. This was achieved by creating larvae expressing just one functioning class of odorant receptor, and hence OSN. Odor response profiles of each OSN class were highly specific and unique. However many OSN-odor pairs yielded variable responses, some of which were statistically indistinguishable from background activity. We used these electrophysiological data, incorporating both responses and spontaneous firing activity, to develop a Bayesian decoding model of olfactory processing. The model was able to accurately predict odor identity from raw OSN responses; prediction accuracy ranged from 12%–77% (mean for all odors 45.2%) but was always significantly above chance (5.6%). However, there was no correlation between prediction accuracy for a given odor and the strength of responses of wild-type larvae to the same odor in a behavioral assay. We also used the model to predict the ability of the code to discriminate between pairs of odors. Some of these predictions were supported in a behavioral discrimination (masking) assay but others were not. We conclude that our model of the peripheral code represents basic features of odor detection and discrimination, yielding insights into the information available to higher processing structures in the brain. Citation: Hoare DJ, Humble J, Jin D, Gilding N, Petersen R, et al. (2011) Modeling Peripheral Olfactory Coding in Drosophila Larvae. PLoS ONE 6(8): e22996. doi:10.1371/journal.pone.0022996 Editor: Bradley Steven Launikonis, University of Queensland, Australia Received March 16, 2011; Accepted July 6, 2011; Published August 9, 2011 Copyright: ß 2011 Hoare et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Funding: Funding was provided by a Biotechnology and Biological Sciences Research Council (http://www.bbsrc.ac.uk) studentship (DJH); a Medical Research Council (http://www.mrc.ac.uk) studentship (JH); The Royal Society (http://royalsociety.org) (CM and MC); Biotechnology and Biological Sciences Research Council grant BB/H009914/1 (CM and MC); and the CARMEN e-science project (Code analysis, repository, and modelling for e-Neuroscience; Engineering and Physical Sciences Research Council (http://www.epsrc.ac.uk) grant EP/E002331/1; RP). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Competing Interests: The authors have declared that no competing interests exist. * E-mail: [email protected]¤ Current address: NIHR National Biomedical Research Unit in Hearing, The University of Nottingham, Nottingham, United Kingdom Introduction In the peripheral olfactory system, odors are represented by a combinatorial code comprising the responses of multiple classes of olfactory sensory neurons (OSNs). Investigation of the contribu- tion of individual OSNs to this code is hampered by complexity; identification of specific OSNs in situ is difficult as most animals possess tens or thousands of cells of each OSN class. In contrast, the olfactory system of the Drosophila larva comprises 21 unique pairs of OSNs, most expressing just a single class of olfactory receptor (OR), and each projecting to its cognate glomerulus in the larval antennal lobe [1–3]. We can record the electrophysiological activity of individual OSNs in vivo, and the larva’s genetic tractability enables analysis of the response profiles of individual, identifiable OSNs expressing specific ORs [4]. This system provides us with the possibility of describing the peripheral olfactory code for a complete OSN population in an intact organism. In a previous study [4], we found that the firing responses of identified larval OSNs to specific pure odors were variable. OSNs of a given class responded reliably to some odors, but not to others. This variability was consistent for specific odor-OSN pairs and was not dependent on odor type or concentration, stimulus duration, genotype or inter-individual differences [4]. Decisively, in larvae expressing only two functional OSNs, one OSN class showed 100% responses to repeated, identical presentations of a given odor, whilst the other OSN class showed variable (,100%) responses to the same presentations of the same odor [4]. Thus, for some odor-OSN pairs, firing responses vary to the extent that they are sometimes statistically indistinguishable from background ‘noise’. The response variability of individual OSNs implies that information reaching the CNS from individual OSNs may be ambiguous. We wished to explore how more reliable coding might emerge at the population level. To address this, we chose a Bayesian decoding approach [5] that would enable us to estimate how much odor identity information can be extracted from OSN activity by downstream neural circuits – in other words, how accurately a target odor can be identified based on the raw peripheral activity alone. First, we exploited the ability to create larvae expressing just one functioning class of OSN to characterise the electrophysiological response profiles of 19 of the 21 OSNs to a panel of 19 biologically-relevant pure odors. This provided quantitative spike PLoS ONE | www.plosone.org 1 August 2011 | Volume 6 | Issue 8 | e22996
11
Embed
Modeling peripheral olfactory coding in drosophila larvae
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Modeling Peripheral Olfactory Coding in DrosophilaLarvaeDerek J. Hoare¤, James Humble, Ding Jin, Niall Gilding, Rasmus Petersen, Matthew Cobb, Catherine
McCrohan*
Faculty of Life Sciences, The University of Manchester, Manchester, United Kingdom
Abstract
The Drosophila larva possesses just 21 unique and identifiable pairs of olfactory sensory neurons (OSNs), enablinginvestigation of the contribution of individual OSN classes to the peripheral olfactory code. We combinedelectrophysiological and computational modeling to explore the nature of the peripheral olfactory code in situ. Werecorded firing responses of 19/21 OSNs to a panel of 19 odors. This was achieved by creating larvae expressing just onefunctioning class of odorant receptor, and hence OSN. Odor response profiles of each OSN class were highly specific andunique. However many OSN-odor pairs yielded variable responses, some of which were statistically indistinguishable frombackground activity. We used these electrophysiological data, incorporating both responses and spontaneous firing activity,to develop a Bayesian decoding model of olfactory processing. The model was able to accurately predict odor identity fromraw OSN responses; prediction accuracy ranged from 12%–77% (mean for all odors 45.2%) but was always significantlyabove chance (5.6%). However, there was no correlation between prediction accuracy for a given odor and the strength ofresponses of wild-type larvae to the same odor in a behavioral assay. We also used the model to predict the ability of thecode to discriminate between pairs of odors. Some of these predictions were supported in a behavioral discrimination(masking) assay but others were not. We conclude that our model of the peripheral code represents basic features of odordetection and discrimination, yielding insights into the information available to higher processing structures in the brain.
Citation: Hoare DJ, Humble J, Jin D, Gilding N, Petersen R, et al. (2011) Modeling Peripheral Olfactory Coding in Drosophila Larvae. PLoS ONE 6(8): e22996.doi:10.1371/journal.pone.0022996
Editor: Bradley Steven Launikonis, University of Queensland, Australia
Received March 16, 2011; Accepted July 6, 2011; Published August 9, 2011
Copyright: � 2011 Hoare et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permitsunrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Funding: Funding was provided by a Biotechnology and Biological Sciences Research Council (http://www.bbsrc.ac.uk) studentship (DJH); a Medical ResearchCouncil (http://www.mrc.ac.uk) studentship (JH); The Royal Society (http://royalsociety.org) (CM and MC); Biotechnology and Biological Sciences Research Councilgrant BB/H009914/1 (CM and MC); and the CARMEN e-science project (Code analysis, repository, and modelling for e-Neuroscience; Engineering and PhysicalSciences Research Council (http://www.epsrc.ac.uk) grant EP/E002331/1; RP). The funders had no role in study design, data collection and analysis, decision topublish, or preparation of the manuscript.
Competing Interests: The authors have declared that no competing interests exist.
no response. There was no significant correlation between mean
response intensity (spikes/s) and the frequency with which a
response above criterion was observed for each odor/OSN
combination (r56 = .245, p = 0.066).
A Bayesian decoding model of the peripheral codeOur consistent finding of considerable variation in responses of
OSNs to specific odors raised the question of whether the
robustness of odor discrimination is increased by integration of
information at the population level. We hypothesized that reliable
information transmission emerges at the ensemble level by
integrating the responses of multiple OSNs. To test this
hypothesis, we constructed a Bayesian decoding model that
integrates the responses of multiple OSNs in a statistically efficient
manner. We used our electrophysiological data from single-
functional-OSN larvae to develop the model (see Methods). We
included the actual spike count for each functional OSN during
1 s stimulation with each odor, regardless of the level of
spontaneous activity. Thus the model incorporated responses as
well as spontaneous (background) activity of both responding and
non-responding OSNs, which all together contribute to the
combinatorial code for a given odor. Four OSNs were excluded
due to shortage of (Or45b), or lack of (Or22c, Or33a and Or82a)
electrophysiological data. One odor (octanal) was excluded
because no single-functional-OSN strain showed an electrophys-
iological response to it. The input to the model on any given trial
therefore consisted of randomly selected samples of the activity of
each of 15 OSNs during presentation of one of 18 target odors.
The corresponding output of the model was a prediction of the
odor most likely to have elicited the input – the model was
effectively required to identify the input as one of the target odors.
The fact that the OSN responses exhibit considerable variability in
their response to a given odor, makes this a demanding 18-
alternative forced choice task; chance level was 5.6%. The model’s
performance for each of the 18 target odors is plotted in Figure 3A.
Despite the considerable variability of the firing responses for most
odor-OSN combinations, every target was significantly correctly
predicted. The mean accuracy of stimulus identification across all
odors was 45.260.2%, eight times greater than chance. However,
not all odors were equally well predicted. The most accurate
prediction level was for 2-heptanone (76.860.3% of trials); the
least accurate was for pentanol (12.360.2%). The model was
particularly efficient at detecting aliphatic esters (ethyl…pentyl
acetate) (range = 49.160.4% to 69.460.3%).
To determine the robustness of the peripheral odor represen-
tation, we examined the effect of progressively eliminating
individual OSNs. At each stage we recomputed the prediction
accuracy based on all 15 subsets of 14 cells and determined the
OSN whose removal produced least performance decrement. This
OSN was then eliminated. This procedure was repeated until only
a single OSN remained (Figure 3B). The first OSN to be removed
was Or33b/47a, which had virtually no effect on the accuracy of
the model (following removal of this OSN, the model’s accuracy
actually increased from 42.2360.08% to 45.2260.09%), suggest-
ing that information from this OSN is not necessary for the
detection of the odors studied here (this OSN responded only to
pentyl acetate, which was detected by five other OSNs). Indeed,
the first five OSNs (Or33b/47a, Or83a, Or74a, Or35a and
Table 1. Summary of electrophysiological activity of single olfactory sensory neurons (OSNs) in Drosophila larvae.
Spontaneous activity (Hz) Firing rate during stimulation (Hz)
Response above criterion Absolute
OSN Mean SE n Min. Max. Mean SE n Min. Max. Min. Max.
Or1a 3.2 0.8 24 0 16 4 0.8 30 0 21 0 37
Or13a 1.5 0.2 24 0 4 7.9 2.1 18 0 30 0 59
Or24a 7.3 0.8 24 0 22 9.6 1.1 46 29 38 0 51
Or30a 3.8 0.3 24 0 6 11.1 1.4 10 0 67 0 74
Or33b/47a 0.7 0.2 24 0 3 10 0.5 8 3 41 2 50
Or35a 7 0.9 24 0 20 16.4 1.4 16 232 109 0 115
Or42a 4 0.8 24 0 17 41.6 1 38 0 133 0 137
Or42b 5.5 1.1 24 0 18 14.8 1.4 12 0 80 0 104
Or45a 2.7 0.8 24 0 8 6.1 0.5 22 0 19 0 28
Or45b 1.9 0.4 24 0 6 23.5 1.4 8 4 45 9 53
Or49a 5.2 0.9 24 0 18 23.2 0.4 21 215 0 0 10
Or59a 1.2 0.3 24 0 4 17.6 1 17 0 61 0 69
Or63a 1.2 0.3 24 0 5 15.9 1 14 0 43 0 49
Or67b 6 0.4 24 0 15 8.1 0.4 36 0 44 1 48
Or74a 4.9 0.5 24 0 12 15.1 0.9 22 28 51 0 60
Or83a 7.9 1.6 24 0 27 14.8 0.8 14 0 33 3 47
Spontaneous activity = the activity of a single OSN in the second prior to each olfactory stimulation. Firing rate = the activity of a single OSN during 1 second stimulationwith one of 19 odors. Min. and max. denote the minimum and maximum rates observed. Response above criterion = change in spike frequency above/below theprobabilistic response criterion (a change of 65 Hz during stimulation as compared to the spontaneous activity seen in that OSN in the 10 s prior to stimulation – seeMaterials and Methods). Absolute = absolute OSN activity during stimulation. Larvae with a single functional OSN were [OrX-Gal4/UAS-Orco ; Orco2/2], constructedfollowing the protocol in [7]. Or33b and Or47a are co-expressed in the same neuron, so their data were pooled. To be certain that spontaneous activity was obtainedfrom a functional OSN, a response to at least one odor had to be detected in that OSN. No responses were detected for Or22c, Or33a and Or82a, so there are no data forthese three classes of OSN.doi:10.1371/journal.pone.0022996.t001
Olfactory Coding in Drosophila Larvae
PLoS ONE | www.plosone.org 4 August 2011 | Volume 6 | Issue 8 | e22996
Or42a) were removed from the model with only a slight decline in
its accuracy (from 42.2360.08% to 40.8460.08%). There was no
significant correlation between the order in which OSNs were
removed from the model and number of odors to which they
responded (r14 = 0.362, p = 0.204), showing that the efficiency of
the model is not just based on the number of odors that each OSN
can detect. Not surprisingly, as OSNs were removed from the
model, certain odors could no longer be predicted at all. For
example, when only Or63a was left, the overall (mean) accuracy of
the model was 14.7360.03%, but this OSN was unable to
correctly identify some odors, such as butanol….nonanol.
Testing the model using behavioral assaysAlthough the model was based on data collected from 19 OSNs,
rather than the full complement of 21, and despite obvious
differences in stimulus duration (1 s for electrophysiology vs 5 min
for behavior), we decided to explore whether the model could
nevertheless make useful predictions about behavioral responses to
odors. We tested responses of wild-type, 21-functional OSN w1118
larvae to individual odors using a mass locomotory assay. The
results are shown in Figure 4. A significant response index – either
attraction or repulsion - was obtained for 15 out of 19 odors. To
explore any relationship between the model’s accuracy in predicting
a target odor (Figure 3A) and the strength of the behavioral response
(response index, either positive or negative) to that odor, we carried
out a correlation analysis (octanal was omitted, since this was not
used to generate the model). There was no correlation between the
two parameters. Some odors that were predicted well by the model
yielded a weak behavioral response (e.g. propyl acetate, 2-
heptanone, ), and vice versa (e.g. pentanol, hexanol).
To further explore the ability of the model to reflect behavior,
we examined how well a Bayesian model could discriminate
Figure 1. Electrophysiological activity of single, identified OSNs during 1 s stimulation (bar) with odors. Recordings are from single-functional-OSN larvae. A. Each trace for a given OR class is from the same larva. The Or35a OSN was activated by butanol and propyl acetate, butshowed no response to anisole or 2-heptanone; the Or49a OSN was inhibited by butanol and 2-heptanone, but showed no response to the other twoodors; the Or59a OSN was activated by anisole, but showed no response to the other three odors. B, C. The Or35a OSN showed variable responses toboth butyl acetate and hexanol. B and C were recorded from separate larvae, each of which showed a response (top trace) or no response (lowertrace) to identical presentations of an odor.doi:10.1371/journal.pone.0022996.g001
Olfactory Coding in Drosophila Larvae
PLoS ONE | www.plosone.org 5 August 2011 | Volume 6 | Issue 8 | e22996
between pairs of odors: that is, when the model’s target was odor
A, how often the OSN activity profile of that odor could be
distinguished from that of odor B. For each odor pair of interest,
we constructed a Bayesian model based on the 15 OSNs identified
above and trained it to discriminate between the two odors. (The
procedures were otherwise identical to those used above). Figure 5
presents a discrimination matrix showing the ability of the model
to discriminate all possible odor pairs. The model was presented
with OSN responses to each pair of odors and required to
discriminate between them. Chance performance was 50%. Every
odor pair was discriminated above this level and, with 11
exceptions, showed a discrimination value of $75%. The lowest
levels of discrimination were generally found between structurally
similar pairs of odors (e.g. pentanol/hexanol – 63–66%). Within
functional groups, the highest levels of discriminability were
detected between hexanoic and nonanoic acid (98%), while the
most consistent discriminability was seen within the four
We tested the output of the discrimination matrix by
studying the behavior of wild-type larvae. We used a ‘masking
test’ [9], in which wild-type larvae were required to detect a test
odor in the presence of a continuous background of the other,
masking, odor. We chose pairs of odors that were either poorly
discriminated by the model (benzyl acetate and hexanol;
pentanol and hexanol; ethyl acetate and heptanal – 63–68%
discriminability; Figure 6A–C) or that were well discriminated
(butyl acetate and octanol; hexanoic acid and hexanol;
hexanoic acid and pentanol – 98–100% discriminability;
Figure 7A–C). If two odors are hard to distinguish, larvae
should find it difficult to detect the test odor against the
background masking odor; the task should be easier for odor
pairs that are easy to distinguish.
In some cases the model was apparently a good predictor of
behavioral discrimination. For example, two of the three odor
pairs that were predicted by the model to be poorly discriminated
- benzyl acetate and hexanol, and pentanol and hexanol - were
also poorly discriminated in the masking test (Figure 6A,B).
However, for the heptanal/ethyl acetate pairing, responses to
each odor were not significantly reduced in the presence of the
other as a mask, demonstrating good behavioral discrimination
(Figure 6C). Behavioral response indices to butyl acetate and
octanol, and to hexanol and hexanoic acid were all significantly
reduced in the presence of the other odor as a mask (Figure 7A,B),
despite the model’s prediction of good discrimination between
these odor pairs. The third pair of odors, predicted to be well
discriminated, yielded asymmetrical data in the masking test. The
response to hexanoic acid was not significantly reduced in the
presence of a pentanol mask, whereas in the reciprocal test the
response to pentanol was significantly reduced in the presence of
hexanoic acid (Figure 7C). These data show that predictions
arising from our model of peripheral ONS coding are not always
correlated with behavioral odor discrimination, and that
discrimination as measured by the masking test is not reciprocal
for every odor pair.
Figure 2. Summary of electrophysiological responses of identified OSNs to a panel of 19 odors. Blue squares indicate inhibition; red tobrown show excitation. Numbers in ‘response reliability’ key indicate the percentage of times that an identified OSN responded to a given odor,rounded up to the nearest 10%. OR33b and OR47a are co-expressed in the same OSN (indicated in blue) and show identical response profiles.doi:10.1371/journal.pone.0022996.g002
Olfactory Coding in Drosophila Larvae
PLoS ONE | www.plosone.org 6 August 2011 | Volume 6 | Issue 8 | e22996
Figure 3. Decoding results of the Bayesian model of peripheral processing. A. The model was presented with responses to each of the 18odors (‘target’) and had to identify which odor induced the response profile. The graph shows the mean percentage of trials (6 SEM) on which themodel correctly identified the target, using 4000 simulations for each odor. Dashed line indicates the percentage expected by chance, grey band
Olfactory Coding in Drosophila Larvae
PLoS ONE | www.plosone.org 7 August 2011 | Volume 6 | Issue 8 | e22996
Discussion
In this study we present the first description of peripheral
olfactory coding in a near-complete (19/21) population of OSNs,
and in an intact organism. We extend our previous finding that
many odor-OSN combinations yield highly variable responses;
based on an objective probabilistic criterion, many ‘responses’ are
not statistically different from spontaneous changes in background
firing activity in the same neuron [4]. Response uncertainty in the
peripheral olfactory system has been reported for other organisms.
Mouse MOR71 cells responded consistently to acetophenone but
showed qualitative response variability to benzaldehyde [11];
similar differences have been reported in MOR23 cells [12]. In
Anopheles mosquitoes, TE1A OSNs respond .80% of the time to
4-ethylphenol, but ,20% of the time to pentanoic acid [13]. In
the vast majority of organisms, where there are many neurons
within each OSN class, quantitative and qualitative variability in
responses provides a continuum of overall response intensity
within the class. However, the Drosophila larva has only a single
pair of OSNs in each class, so must‘cope’ with variability as an
integral part of the peripheral code.
In apparent contrast to our findings, Asahina et al. [14]
recorded odor-evoked calcium signals in OSN axon terminals
within the larval antennal lobe and found that responses were
predictable (invariant) for a given odor-OSN combination.
However, axonal calcium imaging does not provide a complete
reflection of firing activity in sensory dendrites. For example, the
corresponds to P,0.01 confidence limits (see Methods). B. Relative contribution of each OSN class to the accuracy of the model. Data show meanaccuracy levels with a progressive reduction in the number of OSN classes in the model. Standard errors are smaller than the size of the data points.OSN classes contributing least to the ability of the model to accurately predict odors were iteratively removed. The number of odors detected by theremoved OSN class is given underneath each OSN label. Dashed line indicates chance level of correct odor prediction. The X-axis shows the numberof OSNs in the model, with the full model (n = 15) as the first value.doi:10.1371/journal.pone.0022996.g003
Figure 4. Behavioral responses of wild-type w1118 larvae to 19odors. Mean behavioral response indices 6 SEM. Larvae werestimulated with a point source of odor in a mass behavioral test.Response indices were compared with a theoretical value of zero usingone-sample t-tests. * = P#.05; ** = P#.01; ** = P#.001. n = 8 assays perodor.doi:10.1371/journal.pone.0022996.g004
Olfactory Coding in Drosophila Larvae
PLoS ONE | www.plosone.org 8 August 2011 | Volume 6 | Issue 8 | e22996
us to think that the model does indeed reflect important aspects of
sensory processing in this organism.
In creating the model, our aim was to generate a representation
of the peripheral code. However, it was still interesting to explore
its potential to predict behavior. There was no correlation between
the model’s ability to identify a target odor and the behavioral
response index for the same odor. Similarly, when we tested
discrimination between pairs of odors, the model was not a reliable
predictor for behavioral discrimination. These findings were not
surprising, and may have a number of explanations. First, the
model may not adequately reflect the peripheral code owing to the
limitations in our data set referred to earlier. In particular, we were
able to record from only 19 of the 21 larval OSNs. The two
‘missing’ OSNs were present in the 21-functional OSN larvae used
in the masking test, and may be decisive for accurate identification
and discrimination of some or all of these odors. Second, the odor
presentation regimes (timing, concentration) differed between
electrophysiological and behavioral tests and this would be
expected to influence the measured output. Third, behavioral
output reflects the integrative processing of olfactory information
by the brain, whereas the model was based on peripheral activity
alone. The ability to detect and correctly identify a given odor is
necessary, but not sufficient, to elicit a behavioral response to that
odor. The latter also depends on the adaptive and behavioral
relevance of the odor and may present as either attraction,
repulsion, or no behavioral response. Brain processing could also
explain how an odor pair that is poorly discriminated by the
peripheral model (for example, heptanal/ethyl acetate) is much
better discriminated by the whole animal; in this case, discrim-
ination must be sharpened up centrally. In the adult fly brain, odor
detection is sharpened by differential amplification and modula-
tion of signals from OSNs and their cognate glomerulus, together
with fast and rapidly accommodating firing responses in projection
neurons [19]. Lateral inhibition between glomeruli, which
sharpens the signal by increasing the signal:noise ratio, appears
to be particularly important [20]; intraglomerular inhibition may
also play a role [21]. There are similar structures in the larval
antennal lobe [22], and the output of OSNs and glomeruli is
modulated by inhibitory local interneurons and projection
neurons, at least one of which mediates concentration-invariant
odor perception [14]. Such central processing could be used not
only to enhance detection of individual odors but also to improve
discrimination between odors. In the mammalian olfactory bulb,
enhanced cholinergic neurotransmission both sharpens the
olfactory receptive fields of mitral cells and increases behavioral
pairwise odor discrimination [23]. A further observation from the
behavioral masking tests was that reciprocal odor discrimination
could be asymmetrical. In this test, the two odors are presented
differently – one as a background odor and the other a localised
source. The effects of these two kinds of presentation on odor
gradients within the plate could influence the way in which each is
perceived by the brain.
We conclude that, even if the model were deemed to be a fairly
good representation of the basic peripheral code, assuming that its
Figure 5. Ability of the Bayesian model to discriminate between pairs of odors. The model was presented with OSN responses to each pairof odors and required to discriminate between them. The table shows the % of times an odor pair was correctly discriminated; each discriminationwas run through the model 4000 times. Values are not strictly reciprocal on either side of the diagonal because of the sampling method used by themodel. Chance discrimination = 50%.doi:10.1371/journal.pone.0022996.g005
Olfactory Coding in Drosophila Larvae
PLoS ONE | www.plosone.org 9 August 2011 | Volume 6 | Issue 8 | e22996
output is directly translating into odor-induced behavior implies
that important and essential aspects of central processing will be
overlooked. However, the model was able to identify a range of
ecologically relevant odors on the basis of the peripheral responses
they induce, supporting the view that the peripheral code can
perform this task, albeit crudely, without the need for central
Figure 6. Testing the Bayesian model using behavioral discrimination of odor pairs. Masking experiment for pairs of odors predicted bythe model to be poorly discriminated (63–68% discriminability). Control w1118 larvae were tested in a mass olfactory experiment, presented with alocalised odor (‘test’) and a masking odor (‘mask’). For full details, see text. * = P,0.01, *** = P,0.001, n.s. = not significant. n = 8 assays per condition.doi:10.1371/journal.pone.0022996.g006
Figure 7. Testing the Bayesian model using behavioral discrimination of odor pairs. Masking experiment for pairs of odors predicted bythe model to be well discriminated (98–100% discriminability). Control w1118 larvae were tested in a mass olfactory experiment, presented with alocalised odor (‘test’) and a masking odor (‘mask’). For full details, see text. * = P,0.01, *** = P,0.001, n.s. = not significant. n = 8 assays per condition.doi:10.1371/journal.pone.0022996.g007
Olfactory Coding in Drosophila Larvae
PLoS ONE | www.plosone.org 10 August 2011 | Volume 6 | Issue 8 | e22996
processing. An interesting aim for future research will be to
explore further how far the initial olfactory code embodied by our
Bayesian model places constraints on olfactory behaviour, thus
providing insight into the nature and function of central
processing.
Supporting Information
Figure S1 Variable responses for a given odor-OSNcombination are not a function of stimulus flow rate orconcentration. Individual w1118 larvae (1–4) were stimulated
with 2% butanol at three different flow rates (A) or 0.2% butanol
at 30 ml/s (B). In all cases, butanol induced qualitative response
variability; sometimes the OSNs responded, sometimes they did
not. For clarity, responses are ordered in terms of whether there
was a response or not; there was no order effect; stimuli were
presented in random order.
(TIF)
Figure S2 Electrophysiological responses of identifiedlarval OSNs. Larvae from nine single Or strains (Or1a – Or45a)
were stimulated with 19 odors. Responses are given as mean (6
SEM) firing rates of single OSNs above an objective response
criterion. Percentages indicate the proportion of odor presenta-
tions that elicited a response above criterion when stimulated with
a given odor, in preparations in which the functional OSN had
been identified by showing a response to another odor (there are
no percentages for the Or33b OSN, which responded to only a
single odor). n$8 tests per odor/OSN combination. (Data for
Or13a, Or42a and Or42b are taken from [4], Figure 5).
(TIF)
Figure S3 Electrophysiological responses of identifiedlarval OSNs. Larvae from eight single Or strains (Or45b –
Or83a) were stimulated with 19 odors. Responses are given as
mean (6 SEM) firing rates of single OSNs above an objective
response criterion. Percentages indicate the proportion of odor
presentations that elicited a response above criterion when
stimulated with a given odor, in preparations in which the
functional OSN had been identified by showing a response to
another odor (there are no percentages for the Or45b and Or47a
OSNs, which responded to only a single odor). n$8 tests per
odor/OSN combination.
(TIF)
Author Contributions
Conceived and designed the experiments: DJH JH RP MC CM.
Performed the experiments: DJH JH DJ NG. Analyzed the data: DJ JH
RP MC CM. Contributed reagents/materials/analysis tools: RP MC CM.
Wrote the paper: DJ RP MC CM.
References
1. Ramaekers A, Magnenat E, Marin EC, Gendre N, Jefferis GSXE, et al. (2005)Glomerular maps without cellular redundancy at successive levels of the
olfactory representation in mushroom bodies of Drosophila larvae. Proc Natl AcadSci USA 102: 10314–10319.
3. Gomez-Marin A, Duistermars BJ, Frye MA, Louis M (2010) Mechanisms of
odor-tracking: multiple sensors for enhanced perception and behaviour. FrontCell Neurosci 4: 6.
4. Hoare DJ, McCrohan CR, Cobb M (2008) Precise and fuzzy coding by olfactorysensory neurons. J Neurosci 28: 9710–9722.
5. Quian Quiroga R, Panzeri S (2009) Extracting information from neuronal
populations: information theory and decoding approaches. Nat Revs Neurosci10: 173–185.
6. Vosshall LB, Hansson BS (2011) A unified nomenclature system for the insectolfactory co-receptor. Chem Senses 36: 497–498.
7. Fishilevich E, Domingos AI, Asahina K, Naef F, Vosshall LB, et al. (2005)
Chemotaxis behavior mediated by single larval olfactory neurons in Drosophila.Curr Biol 15: 2086–2096.
8. Jan LY, Jan YN (1976) Properties of the larval neuromuscular junction inDrosophila melanogaster. J Physiol 262: 189–214.
9. Kreher SA, Mathew D, Kim J, Carlson JR (2008) Translation of sensory inputinto behavioral output via an olfactory system. Neuron 59: 110–24.
10. Cobb M (1999) What and how do maggots smell? Biol Rev 74: 425–459.
11. Bozza T, Feinstein P, Zheng C, Mombaerts P (2002) Odorant receptorexpression defines functional units in the mouse olfactory system. J Neurosci 22:
3033–3043.12. Grosmaitre X, Vassalli A, Mombaerts P, Shepherd GM, Ma M (2006) Odorant
responses of olfactory sensory neurons expressing the odorant receptor MOR23:
A patch clamp analysis in gene-targeted mice. Proc Natl Acad Sci USA 103:1970–1975.
13. Qiu YT, van Loon JJA, Takken W, Meijerink J, Smid HM (2006) Olfactorycoding in antennal neurons of the malaria mosquito Anopheles gambiae. Chem
Senses 31: 845–863.14. Asahina K, Louis M, Piccinotti S, Vosshall LB (2009) A circuit supporting
concentration-invariant odor perception in Drosophila. J Biol 8: 9.
15. Ignell R, Root CM, Birse RT, Wang JW, Nassel DR, et al. (2009) Presynapticpeptidergic modulation of olfactory receptor neurons in Drosophila. Proc Natl
Acad Sci USA 106: 13070–13075.16. Kreher SA, Kwon AY, Carlson JR (2005) The molecular basis of odor coding in
the Drosophila larva. Neuron 46: 445–456.
17. Guo S, Kim J (2010) Dissecting the molecular mechanism of drosophila odorantreceptors through activity modeling and comparative analysis. Proteins 78:
381–399.18. Hallem EA, Carlson JR (2006) Coding of odors by a receptor repertoire. Cell
125: 143–160.
19. Bhandawat V, Olsen SR, Gouwens NW, Schlief ML, Wilson RI (2007) Sensoryprocessing in the Drosophila antennal lobe increases reliability and separability of
ensemble odor representations. Nat Neurosci 10: 1474–1482.20. Olsen SR, Wilson RI (2008) Lateral presynaptic inhibition mediates gain control
in an olfactory circuit. Nature 452: 956–960.21. Root CM, Masuyama K, Green DS, Enell LE, Nassel DR, et al. (2008) A
presynaptic gain control mechanism fine-tunes olfactory behavior. Neuron 59:
311–321.22. Stocker RF (2008) Design of the larval chemosensory system. Adv Exptl Med
Biol 628: 69–81.23. Chaudhury D, Escanilla O, Linster C (2009) Bulbar acetylcholine enhances
neural and perceptual odor discrimination. J Neurosci 29: 52–60.
Olfactory Coding in Drosophila Larvae
PLoS ONE | www.plosone.org 11 August 2011 | Volume 6 | Issue 8 | e22996